Skip to Main content Skip to Navigation
New interface

Vehicular traffic analysis based on Bluetooth sensors traces

Abstract : The pervasiveness of personal radio devices and the high penetration rate of these technologies in vehicles have, in recent years, made a strong case for the development of new traffic measurement techniques based on the analysis of the radio access network activity levels. In this thesis, we explore the use of sensor data gathered through Bluetooth (BT) passive scanning. Bluetooth sensors provide a cost-effective, low-impact and easy to deploy alternative to conventional techniques. They are adapted for mass deployment in urban areas at relatively low investment and maintenance costs. However, the BT indirect detection process may introduce bias and uncertainties that hinder the accuracy of the derived vehicular traffic metrics. In this context, we investigate the capacity to use Bluetooth sensors as a reliable sole data source for intelligent traffic systems in urban areas. Our work focuses on improving the accuracy of the obtained estimations of the traffic flow and the travel speed. The first part of this work concerns the task of vehicular traffic flow quantification from Bluetooth sensor data. We adopted a data-driven approach relying on statistical and machine learning models. We first considered traffic flow estimation in one sensing pose. Then, we proposed a model for network-scale flow estimation. In this contribution, we also introduced the transfer learning problem required to limit the need to acquire labelled training data for each new deployment. In the second part, we focus on the task of estimating the average travel speed. We propose an algorithm that uses the collected data about the quality of the received signal to improve the matching process and weigh individual vehicle speed contributions in calculating the average speed. During this work, we also developed a simulation framework of BT scanning for vehicular traffic monitoring. The simulator allows us to study and identify the factors impacting the probability, for one sensor, of detecting an active BT connection in its detection range and generate synthetic training datasets to handle data scarcity.
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Friday, December 17, 2021 - 5:44:08 PM
Last modification on : Wednesday, January 26, 2022 - 2:28:16 PM
Long-term archiving on: : Friday, March 18, 2022 - 7:46:47 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03489663, version 1



Safa Boudabous. Vehicular traffic analysis based on Bluetooth sensors traces. Machine Learning [cs.LG]. Institut Polytechnique de Paris, 2021. English. ⟨NNT : 2021IPPAT036⟩. ⟨tel-03489663⟩



Record views


Files downloads